Uncertainty Based Chaotic Pigeon Inspired Optimized Feature Selection for Effective Dyslexia Prediction using Density Peak Clustering

Abstract

The role of cognitive computing in the domain of medical monitoring substantially aids the specialist in detecting concealed disorders at an initial phase. Dyslexia is not a disorder but a long-term problem affecting children's learning and reading abilities. The indicators used for dyslexic detection are vague and uncertain due to their poor quality. Hence, it is essential to examine the importance of features based on the amount of information each attribute can contribute. While all the attributes are involved in prediction of dyslexia, it affects the performance of the prediction model due to high dependency among attributes known as redundancy. Most of the existing feature selection models does not focus on handing uncertainty among attributes. This paper focuses to handle uncertainty by accomplishing clustering of instances and discovering the most relevant features in each cluster class for effective prediction of dyslexic among children. Based on the density of the nearest neighboring instances, the centroids are selected and clustered by developing a density peak clustering algorithm. The features which are highly relevant to predict the dyslexia presence or absence is determined by inducing a nature-inspired algorithm known as Pigeon Inspired Optimization (PIO). Unlike conventional PIO, which selects the population in a random manner, this work utilizes chaotic mapping-based PIO is designed to optimize the feature selection more efficiently for dyslexia prediction. The simulation results validated using support vector machine proved that the proposed Density Peak Clustering based pigeon inspired optimization (DPC-PIO) produced a higher accuracy rate compared to other states of arts algorithms for feature selection in dyslexia prediction.